Many of the tenukis are due to a simple problem that I haven't had time to
fix yet.

 

The old mfgo expert system suggested reasons/goals for moves, each with a
value.  After a move was made and the position evaluated, each reason was
checked to see if the move actually achieved the goal.  For example if the
suggestion was "this move gains 20 points by killing group A", after the
move was played (with quiescence search) the new evaluation checked to see
if group A had actually died.

 

There was a fairly elaborate system to combine the results of the goals to
avoid double counting.

 

In the MCTS search, I use the suggested reasons to bias the search, but I
don't have any way to check the results as the old search did.  So many
goals are double counted.  This causes some of the biases to be way to high,
leading to tenukis.

 

Fixing this is not difficult, but there is a lot of new code to write and
debug, since there are many possible interactions between different goals
(about 200 total goals are checked).

 

David

 

From: [email protected]
[mailto:[email protected]] On Behalf Of Petri Pitkanen
Sent: Thursday, December 30, 2010 12:08 AM
To: [email protected]
Subject: Re: [Computer-go] News on Tromp-Cook ?

 

In last game it did not look like he was taking it easy :) I think Lower
left B was in real trouble. With my crappy tactival skill I thought that W
could have killed 4 black stones and saved his huge corner group. That would
have been swing of 40 points minimum.

So MFoG fights well most of the time then makes easy mistakes. Usually by
making Tenuki. I have some similar feeling playing MAnyFaces  on my laptop.
Every now and then I escape local disasters because something on global
scale seems to be more important - and some time it really is.

 But still huge effort to really challenge European 2d on a even game. Not
something that could have happened 5 years ago. 

2010/12/29 Mark Boon <[email protected]>

On Wed, Dec 29, 2010 at 11:37 AM, Darren Cook <[email protected]> wrote:

> John didn't notice much difference in the better hardware (*) used on
> day 2 (about 20,000 playouts/second, instead of 10,000 I was getting on
> my notebook).

Interesting. I wasn't aware the hardware was better for day 2. But I
did have a distinct feeling the computer was playing better the last
two games. I was even wondering if John had started to take it easy
after the first two easy wins.

Mark

_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

 

_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to